Authors: Rugved Chavan, Gabriel Hyman, Zoraiz Qureshi, Nivetha Jayakumar, William Terrell, Stuart Berr, David Schiff, Megan Wardius, Nathan Fountain, Thomas Muttikkal, Mark Quigg, Miaomiao Zhang, Bijoy Kundu

Published on: February 05, 2024

Impact Score: 8.05

Arxiv code: Arxiv:2402.03414

Summary

• What is new: A novel non-invasive deep learning approach for analyzing dynamic 2-[18F] fluoro-2-deoxy-D-glucose positron emission tomography (dFDG-PET) brain scans, eliminating the need for invasive arterial blood sampling.
• Why this is important: The limited utilization of dFDG-PET for human brain imaging due to the challenge of characterizing a patient-specific blood input function without invasive arterial blood sampling.
• What the research proposes: A deep learning pipeline utilizing a 3D U-Net based ICA-net for internal carotid arteries segmentation and a Recurrent Neural Network based MCIF-net for deriving model-corrected blood input functions with partial volume corrections.
• Results: The pipeline accurately localized seizure onset regions in brain scans, leading to successful treatment outcomes. The ICA-net achieved an average Dice score of 82.18%, and the MCIF-net exhibited a minimal root mean squared error of 0.0052.

Technical Details

Technological frameworks used: 3D U-Net and RNN

Models used: ICA-net for ICA segmentation and MCIF-net for blood input function calculation

Data used: 50 human brain FDG PET datasets

Potential Impact

This research has implications for neuroimaging companies, healthcare providers focusing on neurological conditions, and developers of imaging software and hardware.

Want to implement this idea in a business?

We have generated a startup concept here: NeuroFlow Analytics.